The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context injection (!command syntax) in the SKILL.md file to gather information about the repository state, including the current branch, default base, commit logs, and status of uncommitted or ignored files. These commands are read-only operations that provide the agent with necessary context to execute its primary function and do not involve exfiltration or unsafe command construction.
[COMMAND_EXECUTION]: The skill's instructions outline a sequence of standard Git operations (git fetch, git stash, git checkout, git add). These commands are restricted to the local repository context and are necessary for the skill's stated purpose of branch management. There is no evidence of arbitrary shell execution or unsafe handling of user-supplied arguments.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local file system and Git history (e.g., file paths, branch names, commit messages). While this constitutes an attack surface, the risk is minimal as the skill performs well-defined Git operations and includes rules to prevent destructive actions like overwriting branches or modifying the .gitignore without explicit request.

create-branch